[steph4n-gh]

System1 teaching sessions now support a complete review and recovery lifecycle: freeze a separate independent qualification cohort for an unchanged passing candidate, explicitly adopt the approved artifact, inspect review reasons, and restore retained approved bytes while preserving lessons. The same operations are available in Python, the CLI, and the existing document workspace.

Qualification consumes failed and interrupted cohorts and spends a fixed confidence budget across session attempts. Required qualification is an additional opt-in policy; old development defaults, ridge fitting, core APIs, required dependencies, and saved-skill format remain compatible. Legacy 1.1 sessions reopen without rewriting, with historical approved skills retained on the first new adoption.

Frozen research compared label selection, existing solvers, a cached encoder, and a contrastive projection. Every proposed research capability missed its prospective gates and is omitted. The new sentiment lifecycle trial failed before qualification; a separate historical banking replay verified failed-candidate refusal, adoption, reopening, and exact rollback. Protocol commits, all misses, source restrictions, invalid-cache evidence, and costs remain in Git. Raw corpora and model weights do not enter the distributions.

Validation:

• 1,387 local tests passed with 22 optional skips on Python 3.12 and independently on native macOS Python 3.13, including 50 focused qualification/recovery contracts.
• Existing offline quality and recorded-Jev regression passed for all three bounded demonstrations.
• Five research isolation/cache checks passed; the teaching workspace was exercised in the browser.
• Keep oversized HTTP boundary inputs/assertions while shortening verbose test identifiers; make numeric-loopback HTTP test startup independent of reverse DNS. Focused workspace checks pass.
• Documentation: 102 Markdown files, 704 local links, nine current release labels; lockfile and whitespace checks passed.
• Separate minimal wheel install outside the checkout passed actual published-1.1.0 session migration without rewriting, exact prediction parity, new adoption, byte-exact legacy rollback, required qualification, qualified rollback, CLI, reflex, and proto assets; SciPy absent.
• Built wheel 206,120 bytes and source archive 9,580,526 bytes; archive-content checks and strict metadata validation passed.

Release notes: docs/releases/1.2.0.md. Existing full-scope assistant and banking gates remain unchanged and unqualified. Sessions remain local single-writer workspaces; independence, representativeness, and supplied-label correctness cannot be established by software.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.